Skip to content

Commit

Permalink
Fix ima for split-kv kernel (#20)
Browse files Browse the repository at this point in the history
  • Loading branch information
bfontain authored Sep 5, 2024
1 parent 582eb8f commit 0c2fb25
Showing 1 changed file with 24 additions and 13 deletions.
37 changes: 24 additions & 13 deletions csrc/flash_attn/flash_api.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -209,7 +209,7 @@ inline int num_splits_heuristic(int batch_nheads_mblocks, int num_SMs, int num_n
return 1;
}

void set_params_splitkv(Flash_fwd_params &params, const int batch_size,
std::tuple<at::Tensor, at::Tensor> set_params_splitkv(Flash_fwd_params &params, const int batch_size,
const int num_heads, const int head_size, const int max_seqlen_k, const int max_seqlen_q,
const int head_size_rounded, const float p_dropout,
const int num_splits, cudaDeviceProp *dprops, struct c10::TensorOptions opts) {
Expand All @@ -221,19 +221,24 @@ void set_params_splitkv(Flash_fwd_params &params, const int batch_size,
// In any case we don't expect seqlen_q to be larger than 64 for inference.
const int num_m_blocks = (max_seqlen_q + 64 - 1) / 64;
params.num_splits = num_splits;
at::Tensor softmax_lse_accum;
at::Tensor out_accum;

if (p_dropout == 0.0f) { // SplitKV is not implemented for dropout
if (num_splits < 1) {
// We multiply number of SMs by 2 to hard-code the fact that we're using 128 threads per block.
params.num_splits = num_splits_heuristic(batch_size * num_heads * num_m_blocks, dprops->multiProcessorCount * 2, num_n_blocks, 128);
}
if (params.num_splits > 1) {
at::Tensor softmax_lse_accum = torch::empty({params.num_splits, batch_size, num_heads, max_seqlen_q}, opts.dtype(at::kFloat));
at::Tensor out_accum = torch::empty({params.num_splits, batch_size, num_heads, max_seqlen_q, head_size_rounded}, opts.dtype(at::kFloat));
softmax_lse_accum = torch::empty({params.num_splits, batch_size, num_heads, max_seqlen_q}, opts.dtype(at::kFloat));
out_accum = torch::empty({params.num_splits, batch_size, num_heads, max_seqlen_q, head_size_rounded}, opts.dtype(at::kFloat));
params.softmax_lseaccum_ptr = softmax_lse_accum.data_ptr();
params.oaccum_ptr = out_accum.data_ptr();
}
TORCH_CHECK(params.num_splits <= 128, "num_splits > 128 not supported");
}

return std::make_tuple(softmax_lse_accum, out_accum);
}

void set_params_alibi(Flash_fwd_params &params, c10::optional<at::Tensor> &alibi_slopes_, int batch_size, int num_heads){
Expand Down Expand Up @@ -394,10 +399,11 @@ mha_fwd(at::Tensor &q, // batch_size x seqlen_q x num_heads x head_size
softcap
);


set_params_splitkv(params, batch_size, num_heads,
head_size, seqlen_k, seqlen_q,
head_size_rounded, p_dropout, /*num_splits*/0, dprops, opts);
// Keep references to these tensors to extend their lifetime
at::Tensor softmax_lse_accum, out_accum;
std::tie(softmax_lse_accum, out_accum) = set_params_splitkv(
params, batch_size, num_heads, head_size, seqlen_k, seqlen_q,
head_size_rounded, p_dropout, /*num_splits*/ 0, dprops, opts);

// number of times random will be generated per thread, to offset philox counter in thc random
// state
Expand Down Expand Up @@ -642,11 +648,14 @@ mha_varlen_fwd(at::Tensor &q, // total_q x num_heads x head_size, total_q := \s
params.v_batch_stride = v_padded.stride(0);
}
params.page_block_size = page_block_size;
// Keep references to these tensors to extend their lifetime
at::Tensor softmax_lse_accum, out_accum;
if (seqlenq_ngroups_swapped) {
// Only apply split-k for decoding
set_params_splitkv(params, batch_size, num_heads,
head_size, max_seqlen_k, max_seqlen_q,
head_size_rounded, p_dropout, /*num_splits*/0, dprops, opts);
std::tie(softmax_lse_accum, out_accum) =
set_params_splitkv(params, batch_size, num_heads, head_size,
max_seqlen_k, max_seqlen_q, head_size_rounded,
p_dropout, /*num_splits*/ 0, dprops, opts);
}

// number of times random will be generated per thread, to offset philox counter in thc random
Expand Down Expand Up @@ -936,9 +945,11 @@ mha_fwd_kvcache(at::Tensor &q, // batch_size x seqlen_q x num_he
params.cache_batch_idx = reinterpret_cast<int *>(cache_batch_idx.data_ptr());
}

set_params_splitkv(params, batch_size, num_heads,
head_size, seqlen_k, seqlen_q,
head_size_rounded, /*dropout*/0.f, num_splits, dprops, opts);
// Keep references to these tensors to extend their lifetime
at::Tensor softmax_lse_accum, out_accum;
std::tie(softmax_lse_accum, out_accum) = set_params_splitkv(
params, batch_size, num_heads, head_size, seqlen_k, seqlen_q,
head_size_rounded, /*dropout*/ 0.f, num_splits, dprops, opts);

if (paged_KV) {
params.block_table = block_table.data_ptr<int>();
Expand Down

0 comments on commit 0c2fb25

Please sign in to comment.